Machine Learning Optimization using Bat Algorithm to Classify Sentiment of Twitter Users

Social-media is a very effective communication media in today’s digital era. Twitter is one of them which widely used by Internet users. Huge number of tweets has encouraged research in the field of text mining, especially in sentiment analysis. Most of sentiment analysis researches which mined data in Bahasa used TF-IDF to assign weight on every word in corpus. This traditional method resulted low accuracy when tested using machine learning methods. In this study, instead of using TF-IDF, we implemented Bat Algorithm to weight every word in corpus. We tested this on Naïve Bayes, Decision Tree and K-NN methods. The result of this study shows that Naïve Bayes, Decision Tree and K-NN methods which classified data weighted using TF-IDF reached accuracy 33.58%, 32.82% and 33.61%, respectively. Afterwards, words in corpus were weighted using Bat Algorithm and tested using the same methods. The test result shows that Naïve Bayes, Decision Tree and K-NN methods reached 39.01%, 76.63% and 66.15% in respectively. It can be inferred that Bat Algorithm usage for weighting words in corpus improves machine learning algorithms to classify sentiment of Twitter users. Moreover, it can be identified that the biggest improvement occurred in Decision Tree algorithm which increased 43.81% accuracy. On the other hand, improvement in Naïve Bayes algorithm is still minor compared to other machine learning algorithms.